Steps to reproduce > the same prompt that was working and retrieving fine from my knowledge file, now seems to hallucinate mixing up results from my knowledge file and generic ChatGPT answers.
If OpenAI is reading this, this is REALLY REALLY PAINFUL that every couple of weeks ago, custom GPTs are not stable. We spend days and days trying to reverse engineer how to best programme customGPTs to make them stable and suddenly we fall back into unpredictable responses.